[
https://issues.apache.org/jira/browse/HBASE-5954?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14222321#comment-14222321
]
Lars Hofhansl commented on HBASE-5954:
--------------------------------------
Well... We do have the APIs already. We do not have a "sync always means hsync"
flag, that would have been easier, but less flexible.
We could remove the API again (as FSYNC_WAL is identical to SYNC_WAL right now
anyway, see the Durability class)... I.e. deprecating FSYNC_WAL.
The idea was that the application knows best. It might do ASYNC_WAL most of the
time, then issue a SYNC_WAL, and occasionally an FSYNC_WAL when it knows that
this is necessary.
Thinking about it, it would nice to get rid of this complexity (and I would no
longer hate this patch either :) ). This needs to be a careful decision,
though. If we want to avoid the mixed handling we could not allow this on a per
column family or even table level, it would need be on or off per cluster. In
earlier discussions we wanted to avoid this due to the high cost of an fsync
(until we have storage classes in HDFS or SSDs everywhere). I'd be in favor of
having this discussion again.
BTW. My patch already has hsync for META always :)
> Allow proper fsync support for HBase
> ------------------------------------
>
> Key: HBASE-5954
> URL: https://issues.apache.org/jira/browse/HBASE-5954
> Project: HBase
> Issue Type: Improvement
> Components: HFile, wal
> Reporter: Lars Hofhansl
> Assignee: Lars Hofhansl
> Priority: Critical
> Fix For: 2.0.0
>
> Attachments: 5954-WIP-trunk.txt, 5954-trunk-hdfs-trunk-v2.txt,
> 5954-trunk-hdfs-trunk-v3.txt, 5954-trunk-hdfs-trunk-v4.txt,
> 5954-trunk-hdfs-trunk-v5.txt, 5954-trunk-hdfs-trunk-v6.txt,
> 5954-trunk-hdfs-trunk.txt, hbase-hdfs-744.txt
>
>
> At least get recommendation into 0.96 doc and some numbers running w/ this
> hdfs feature enabled.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
